异构边缘计算环境下异步联邦学习的节点分组与分时调度策略

为了克服异构边缘计算环境下联邦学习的3个关键挑战,边缘异构性、非独立同分布数据及通信资源约束,提出了一种分组异步联邦学习（FedGA）机制,将边缘节点分为多个组,各个分组间通过异步方式与全局模型聚合进行全局更新,每个分组内部节点通过分时方式与参数服务器通信。理论分析建立了FedGA的收敛界与分组间数据分布之间的定量关系。针对分组内节点的通信提出了分时调度策略魔镜法（MMM）优化模型单轮更新的完成时间。基于FedGA的理论分析和MMM,设计了一种有效的分组算法来最小化整体训练的完成时间。实验结果表明,FedGA和MMM相对于现有最先进的方法能降低30.1%～87.4%的模型训练时间。     

MIMO雷达分组正交波形集优化设计方法

在现代电子对抗中,数字射频存储(DRFM)设备能够快速截获机载脉冲多普勒雷达信号,能够实现对多输入多输出(MIMO)雷达的干扰。MIMO雷达可基于多组相互正交的波形集来对抗DRFM干扰。同时,为最大化MIMO雷达波形分集增益,每个脉冲内发射的波形也需要正交。为了平衡组内和组间的正交性,文中建立了一种分组正交波形集优化模型,其目标函数为组内和组间相关函数性能评估指标值的加权和;为了求解该优化问题,提出了一种分组正交波形集设计方法。所提方法将原优化问题简化为p-范数优化问题,基于MM算法导出了最小化目标函数的迭代求解表达式。仿真结果表明,所提方法可通过改变权重系数来灵活平衡MIMO雷达的干扰抑制性能和距离压缩性能。     

基于资源的混合式学习教学设计

混合式学习模式通过对各种教学手段和方法以及各种学习资源的优化和整合,能够使学生在同一学习过程或周期中达到更好的学习效果,符合了计算机基础教育课程体系和教学模式改革的目的和初衷。本文探讨了如何将混合式学习模式融入到中学课堂教学中,并以高中信息技术课程中的一节课《用photoshop软件为任务替换背景》为例,设计了具体的教学方案,为教学质量和效率的提高提供了新的手段。     

基于正样本对比与掩蔽重建的自监督语音表示学习

针对现有基于对比预测的自监督语音表示学习方法在训练时需要构建大量负样本,其学习效果依赖于大批次训练,需要耗费大量计算资源的问题,提出了一种仅使用正样本进行语音对比学习的方法,并将其与掩蔽重建任务相结合得到一种多任务自监督语音表示学习方法,在降低训练复杂度的同时提高语音表示学习的性能。其中,正样本对比学习任务,借鉴图像自监督表示学习中SimSiam方法的思想,采用孪生网络架构对原始语音信号进行两次数据增强,并使用相同的编码器进行处理,将一个分支经过一个前向网络,另一个分支使用梯度停止策略,调整模型参数以最大化2个分支输出的相似度。整个训练过程中不需要构造负样本,可采用小批次进行训练,大幅提高了学习效率。使用LibriSpeech语料库进行自监督表示学习,并在多种下游任务中进行微调测试,对比实验表明,所提方法得到的模型在多个任务中均达到或者超过了现有主流语音表示学习模型的性能。     

多种群协方差学习差分进化算法

种群多样性与交叉算子在差分进化(DE)算法求解全局优化问题中具有重要作用,该文提出一种多种群协方差学习差分进化(MCDE)算法。首先,采用多种群机制的种群结构,利用每一子种群结合相应的变异策略保证进化过程个体多样性。然后,通过种群间的协方差学习,为交叉操作建立一个适当旋转的坐标系统;同时,使用自适应控制参数来平衡种群的勘测与收敛能力。最后,在单峰函数、多峰函数、偏移函数和高维函数的25个基准测试函数上进行测试,并同其他先进的进化算法对比,实验结果表明该文算法相较于其他算法在求解全局优化问题上达到最优效果。     

基于LDA-LSTM模型在线课程评论情感分析研究

终身化学习背景下,MOOC作为普及性在线学习形式已受到学术界的日益关注。同时,MOOC课程质量与学习者满意度问题亟待解决。研究基于理性选择理论与联通主义理论构建LDA-LSTM深度主题情感分析模型,进而挖掘学习者理性因素与情感极性。实验结果表明,学习者考虑的因素主要具备全面性与多样性的特点;学习者对教师与学习效果普遍给予肯定评价,较少负面评价则针对教师授课风格、课程资源与平台服务质量。研究据此给出了建议策略。     

一种基于优化引导的无线联邦学习异步训练机制

针对在数据异构和资源异构的无线网络中联邦学习训练效率低及训练能耗高的问题,面向图像识别任务,提出了基于优化引导的异步联邦学习算法AFedGuide。利用较高样本多样性的客户端模型的引导作用,提高单轮聚合有效性。采用基于训练状态的模型增量异步更新机制,提高模型更新实时性以及信息整合能力。设计基于模型差异性的训练决策,修正优化方向。仿真结果显示,相较于对比算法,AFedGuide的训练时长平均减少67.78%,系统能耗平均节省65.49%,客户端的准确率方差平均减少25.5%,说明在客户端数据异构和资源异构的无线网络下,AFedGuide可以在较短的训练时间内以较小的训练能耗完成训练目标,并维持较高的训练公平性和模型适用性。     

大数据环境中简化粒子群算法的改进研究

针对大数据存在的高维、强约束和多目标等复杂优化问题,本文提出一种改进的群智能优化算法——狮群简化粒子群算法(LSA-SPSO).该算法将狮群算法的分组思想融入简化粒子群优化算法中,将粒子分为三组寻优,每组使用不同的学习因子和学习维度向量,以此帮助种群执行不同的搜索机制,从而增强了种群的多样性.此外,引入种群育种,有利于粒子跳出局部最优位置,提高了算法的全局搜索性能.仿真实验表明,本文提出的改进算法有效改善了传统群智能算法中存在的不足,可以更好的应用到大数据中.     

基于模糊C均值算法的协作学习分组系统研究

在分析"WebCL系统模型"基础上,根据学习者的学习风格和特征的模糊性,采用模糊C均值(FCM)算法进行分组.实验结果表明,该方法增强了协作学习分组的可靠性,同时能够明显提高协作学习的效果.     

基于特征关系依赖网络的小样本学习方法

小样本学习任务旨在仅提供少量训练样本的情况下完成对测试样本的正确分类.基于度量学习的小样本学习方法通过将样本映射到嵌入空间中,计算样本间距离得到相似性度量以预测类别,但仅对样本特征进行独立映射,而忽略了对整个任务的观察,同时在小样本场景下通过传统方法计算的原型与期望原型存在偏差,导致在查询集上泛化性较低.针对上述问题,提出了特征关系依赖网络(FRDN).特征关系依赖网络包含两个模块:首先使用关系挖掘模块充分挖掘任务中样本的类内与类间关系,将其作为自注意力值对类簇进行调整,以获得判别性更高的任务自适应嵌入空间,计算初始原型;随后使用偏差抑制模块对初始原型进行校正,得到在查询集上泛化性更高的优化原型,进一步提高模型的分类准确率.在MiniImagenet数据集上,该方法1-shot分类准确率59.17%,5-shot准确率74.11%,分别超过传统度量学习方法6.13%与2.83%;在CUB数据集上分别提升9.3%和2.74%.     

Group-sensitive multiple kernel learning for object recognition

In this paper, a group-sensitive multiple kernel learning (GS-MKL) method is proposed for object recognition to accommodate the intraclass diversity and the interclass correlation. By introducing the "group" between the object category and individual images as an intermediate representation, GS-MKL attempts to learn group-sensitive multikernel combinations together with the associated classifier. For each object category, the image corpus from the same category is partitioned into groups. Images with similar appearance are partitioned into the same group, which corresponds to the subcategory of the object category. Accordingly, intraclass diversity can be represented by the set of groups from the same category but with diverse appearances; interclass correlation can be represented by the correlation between groups from different categories. GS-MKL provides a tractable solution to adapt multikernel combination to local data distribution and to seek a tradeoff between capturing the diversity and keeping the invariance for each object category. Different from the simple hybrid grouping strategy that solves sample grouping and GS-MKL training independently, two sample grouping strategies are proposed to integrate sample grouping and GS-MKL training. The first one is a looping hybrid grouping method, where a global kernel clustering method and GS-MKL interact with each other by sharing group-sensitive multikernel combination. The second one is a dynamic divisive grouping method, where a hierarchical kernel-based grouping process interacts with GS-MKL. Experimental results show that performance of GS-MKL does not significantly vary with different grouping strategies, but the looping hybrid grouping method produces slightly better results. On four challenging data sets, our proposed method has achieved encouraging performance comparable to the state-of-the-art and outperformed several existing MKL methods.     

基于改进型深度学习的流量预测

为了解决无线网络中流量的预测精度不高的问题,提出了一种自适应分组的栈式自编码( AG-SAEs)深度学习预测方法。在数据的预处理过程中,首先使用最大最小方式对数据进行归一化处理,并提出一种新型的自适应分组方法,把归一化后的链路数据进行关联性分组;然后,基于深度学习方法建立了一个多输入多输出的预测模型,并将分组后的数据输入到预测模型中,对该模型进行训练来建立输入和输出流量之间的映射关系;最后,为了进一步提高预测精度,在模型的训练过程中,使用改进型的牛顿法来进行权值参数更新。仿真实验以及和其他算法对比的结果证实了所提方案具有更小的预测相对误差。     

基于SPOC的混合式学习的学习分析与应用

在自主开发的个性化教学平台基础上,探索了课堂教学与网络MOOC教学相融合的混合学习模式.根据学习分析结果,有针对性地对学生的学习过程进行干预,并发现教学中的问题,进而不断改善混合学习的教学设计与实施.研究表明, 从群体的行为数据均值考虑,学生的在线学习表现和学生成绩是正相关的,网上MOOC学习对学习效果的提升有较大促进作用.     

基于Leader-Follower的分群移动控制算法

分布式智能体群体移动控制中,简洁高效的多任务分群和群体移动是实现复杂群体运动控制的基础.群体运动会受到各节点运动不统一的影响而产生震荡和波动,使群体的移动距离增加,降低了群体移动效率.受"阻尼减震"的启发,本文对领导者最大速度进行限制,同时调节合力组成比例,对节点间无序力进行弱化,减小无序力对群体移动的负面影响,突出移...     

基于迁移学习的红外图像多目标检测技术

针对用传统方法难以解决城市背景下红外图像多目标检测的问题,采用迁移学习技术把深度学习中可见光域的目标检测框架迁移到红外域中。利用该方法建立的模型的小目标检测性能非常好,在制作的测试集上平均精度mAP（IoU=0.50）为0.858。还对训练数据与模型检测性能之间的关系进行了初步研究。制作了大数据量和小数据量2个训练集,对模型进行训练,然后在相同的测试集上进行测试。通过小数据量训练的模型在制作的测试集上的平均精度mAP(IoU=0.50)为0.615。实验结果表明,数据的多样性、数量、质量等都会影响模型的好坏。     

WIBRO系统中基于SM-STBC的自适应天线分组算法

在多天线系统中,采用空间复用空时块码(STBC,space-time block code)可以在空间复用增益和分集增益之间达到很好的性能折中.文中提出一种应用在无线宽带系统中,基于空间复用STBC的自适应天线分组算法.该算法可以自适应地将天线进行分组,在组内采用STBC分集方案,组间采用空间复用方案.优化分组既能在每个子载波上基于最小SNR最大化准则来进行,也可在整个频带上使用平均互信息最大化准则来确定.仿真结果表明,我们提出的两种自适应天线分组方法都可以有效地改善系统的性能.     

Multi-task learning with deformable convolution

Multi-task learning aims to tackle various tasks with branched feature sharing architectures. Considering its diversity and complexity, discriminative feature representations need to be extracted for each individual task. Fixed geometric structures as a limitation of convolutional neural networks (CNNs) in building models, is also exists and poses a severe challenge in multi-task learning since the geometric variations will augment when we deal with multiple tasks. In this paper, we go beyond these limitations and propose a novel multi-task network by introducing the deformable convolution. Our design, the Deformable Multi-Task Network (DMTN), starts with a single shared network for constructing a shared feature pool. Then, we present task-specific deformable modules to extract discriminative features to be tailored for each task from the shared feature pool. The task-specific deformable modules utilize two new parts, deformable part and alignment part, to extract more discriminative task-specific features while greatly enhancing the transformation modeling capability. Experiments conducted on various multi-task learning types demonstrate the effectiveness of the proposed method. On multiple classification tasks, semantic segmentation and depth estimation tasks, our DMTN exceeds state-of-the-art approaches against strong baselines.     

Avoiding collaborative paradox in multi-agent reinforcement learning

The collaboration productively interacting between multi-agents has become an emerging issue in real-world applications. In reinforcement learning, multi-agent environments present challenges beyond tractable issues in single-agent settings. This collaborative environment has the following highly complex attributes: sparse rewards for task completion, limited communications between each other, and only partial observations. In particular, adjustments in an agent's action policy result in a nonstationary environment from the other agent's perspective, which causes high variance in the learned policies and prevents the direct use of reinforcement learning approaches. Unexpected social loafing caused by high dispersion makes it difficult for all agents to succeed in collaborative tasks. Therefore, we address a paradox caused by the social loafing to significantly reduce total returns after a certain timestep of multi-agent reinforcement learning. We further demonstrate that the collaborative paradox in multi-agent environments can be avoided by our proposed effective early stop method leveraging a metric for social loafing.     

Adaptively Grouped Multilevel Space-Time Trellis Codes

Grouped multilevel space-time trellis codes (GMLSTTCs) utilize multilevel coding (MLC), antenna grouping and space time trellis codes (STTCs) for simultaneously providing coding gain, diversity improvement and increased spectral efficiency. The performance of GMLSTTCs is limited due to predefining of the antenna groups. It has been shown that when perfect or partial channel state information is available at the transmitter, the performance and capacity of space-time coded system can be further improved. In this paper, we present a new code designed by combining MLC, STTCs, antenna grouping and channel state information at transmitter, henceforth referred to as adaptively grouped multilevel space time trellis codes (AGMLSTTCs). AGMLSTTCs use a single full-diversity STTC at initial some levels and multiple STTCs at some later levels. The single full diversity STTC at each initial level spans all transmit antennas and the STTC at each later level spans a group of transmit antennas. The channel state information at the transmitter is used to adaptively group the transmit antennas for the later levels. Instantaneous channel power gain is calculated between each transmit antenna and all the receive antennas. A subset of transmit antennas having maximum channel power gain is selected to form a group. The simulation results show that AGMLSTTCs enable to transmit more than one data symbol per time slot with improved error performance over GMLSTTCs with predefined transmit antenna grouping.     

MapReduce optimization algorithm based on machine learning in heterogeneous cloud environment

We present an approach to optimize the MapReduce architecture, which could make heterogeneous cloud environment more stable and efficient. Fundamentally different from previous methods, our approach introduces the machine learning technique into MapReduce framework, and dynamically improve MapReduce algorithm according to the statistics result of machine learning. There are three main aspects： learning machine performance, reduce task assignment algorithm based on learning result, and speculative execution optimization mechanism. Furthermore, there are two important features in our approach. First, the MapReduce framework can obtain nodes＇ performance values in the cluster through machine learning module. And machine learning module will daily calibrate nodes＇ performance values to make an accurate assessment of cluster performance. Second, with the optimization of tasks assignment algorithm, we can maximize the performance of heterogeneous clusters. According to our evaluation result, the cluster performance could have 19% improvement in current heterogeneous cloud environment, and the stability of cluster has greatly enhanced.